Make sure your generative AI use is VALID.
Whether you love ChatGPT, Bard, and similar generative AI tools, or you think they are a prelude to the Robot Apocalypse, the expression “you can’t put the genie back in the bottle” seems appropriate. As I previously blogged about, I embrace the vast potential of AI to advance human welfare when used wisely. Further, to avoid using it completely puts your team and your company at a disadvantage. I believe this also holds true for generative AI (GAI).
The rapid adoption of AI-empowered chatbots has spawned an uproar about real and imagined calamities that tend to drown out the many benefits this technology affords the users who apply it with integrity and discretion. That said, individuals and businesses would do well to consider, anticipate, and mitigate the riskier aspects of GAI use by adopting a few prudent habits.
Here are a some of the guardrails my team recommends — and yes, I’ve come up with an acronym to make them easier to remember: VALID
V —Verify all responses you receive from ChatGPT by cross-checking what’s reported with independent sources. AI software is super smart about curating, cutting, and pasting info that’s floating around out there in the ether. It is less discerning regarding what’s true, biased, or invented.
A — Avoid using any generative AI services to record or transcribe conference calls, generate email responses, etc. This is the business equivalent of leaving your wallet on the dashboard of your unlocked car with the key in it. However, if you have strong security measures in place, and everyone on the call is an employee operating behind the same firewall (ensuring that the data will remain secure) then GAI can boost productivity, and even remind you to follow up on action items. The key is making sure everyone understands how to tightly control risk while maximizing reward.
L — Be mindful of the LEGAL implications of using ChatGPT to generate copy. It will not have your back when you are sued for copyright infringement or fired for plagiarism. It will not defend you when you inadvertently undermine brand trust by spreading falsehoods. Always check that information provided by generative AI is not only accurate but in the public domain.
I – Protect your INFORMATION. You can put yourself, clients, employers, and all your stakeholders at risk by uploading proprietary information/data to ChatGPT and hoping it will edit or improve your work. It might do that. But it will also serve your information back up to the next person who probes a similar line of inquiry. A rule of thumb is, don’t upload anything that you wouldn’t post on social media. Better yet, just develop a stream of inquiry where you ask the questions and let the chatbot do the talking.
D — Exercise DOUBT. You don’t have to use ChatGPT yourself to become its victim. Cyber criminals are already using ChatGPT and similar software to enhance the quality of their attacks. By now you’ve heard about malicious characters using AI to impersonate a loved one’s voice in order to extort money. They are also using technology to develop realistic phishing campaigns, malware, ransomware, data theft, and other fraudulent activities. If you receive a message asking you to do something urgently and out of the ordinary — even from someone you trust — don’t respond. Instead, reach out to the individual separately, using a different communication channel, and ask them to confirm or deny the request. Either way, they’ll thank you.
One of the things I continue to find unsettling is that generative AI, as with any form of searchable media, makes it easier for people to indulge in confirmation bias. If I ask my favorite AI language model to find evidence for my pet conspiracy theories, it will try to accommodate me. Fortunately, data scientists are drawing attention to the issue of unintended bias in ChatGPT and other bots. Unfortunately, that won’t keep unethical people from developing their own malignantly skewed bots.
As always, our best defense is to use artificial intelligence as a tool to bolster and stimulate our actual intelligence. When chess masters started playing against computers, the computers learned how to beat them consistently. But when the chess master teamed up with a computer to play another computer, the combination was unbeatable. The lesson is to give your AI-powered technology a seat at the table. Collaborate. Ultimately, the best way to ensure that generative AI bots deliver valid results is to deploy that supercomputer sitting inside your head. Think!
Follow me on LinkedIn.